-
Notifications
You must be signed in to change notification settings - Fork 21.4k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Store autocast_gpu_dtype
in custom_fwd
and custom_bwd
for BFloat16 autocast
#88029
Conversation
Signed-off-by: Masaki Kozuki <mkozuki@nvidia.com>
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/88029
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 4b580fc: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks good, thanks for the test cleanup!
Signed-off-by: Masaki Kozuki <mkozuki@nvidia.com>
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
…t16 autocast (pytorch#88029) As per pytorch#87979, `custom_bwd` seems to forcefully use `torch.float16` for `torch.autograd.Function.backward` regardless of the `dtype` used in the forward. Changes: - store the `dtype` in `args[0]` - update tests to confirm the dtype of intermediate result tensors that are outputs of autocast compatible `torch` functions cc @ptrblck @ngimel Pull Request resolved: pytorch#88029 Approved by: https://github.com/ngimel
…t16 autocast (pytorch#88029) As per pytorch#87979, `custom_bwd` seems to forcefully use `torch.float16` for `torch.autograd.Function.backward` regardless of the `dtype` used in the forward. Changes: - store the `dtype` in `args[0]` - update tests to confirm the dtype of intermediate result tensors that are outputs of autocast compatible `torch` functions cc @ptrblck @ngimel Pull Request resolved: pytorch#88029 Approved by: https://github.com/ngimel
As per #87979,
custom_bwd
seems to forcefully usetorch.float16
fortorch.autograd.Function.backward
regardless of thedtype
used in the forward.Changes:
dtype
inargs[0]
torch
functionscc @ptrblck @ngimel